The artificial intelligence boom has operated under a core assumption: larger models are inherently more powerful, and the most powerful models dominate the market. However, the industry is now on the verge of discovering what happens when that assumption begins to falter. Rising costs have already forced users to reconsider smaller, more affordable models. This newfound cost-conscious approach to model selection is unprecedented, and its potential impact on the industry remains uncertain, but it is likely to be profound.

One compelling prediction, articulated by Coinbase co-founder Brian Armstrong, suggests that the majority of tasks will migrate to cheaper models. "Demand for intelligence is near infinite, but 80% of workloads will be running on 99% cheaper models within 12-18 months," Armstrong wrote on X. "20% of workloads will still run on latest gen models where IQ maxing is important." If Armstrong's forecast proves accurate, the shift for the AI industry would be monumental. Historically, most AI companies have competed on quality, defaulting to the most advanced available model. If those same tasks can be effectively handled by cheaper models without compromising quality, it would fundamentally alter the economics of AI.

Critically, much of the savings would come at the expense of major labs, delivering a financial blow to companies like OpenAI and Anthropic just as they approach their IPOs. This represents a potentially seismic industry change, hinging on a single question: Are companies ready to adopt smaller models? Initial tests suggest that, with the right system design, cheaper models can substitute without any loss in quality. In a recent test by legal AI tool Harvey, the company reduced inference costs by three times while maintaining quality. The test, conducted in partnership with inference platform Fireworks AI, combined Claude Opus and Fireworks' GLM 5.1, switching to Opus only for the most demanding tasks. The result was a significantly lower server load and overall cost. "However, the definition of quality is evolving from simply using the most powerful model for everything, to using the best model that gets the right answer most efficiently."

This trend is often framed as a battle between major labs and Chinese models or open-weight alternatives, but that perspective misses the larger point. The real divide is not between proprietary and open models; it is between large models and small ones. You can save money by switching from GPT-5.5 to DeepSeek's V4 Flash, but switching to GPT-5.4-mini works just as well. An active price war is underway between in-house inference from major labs and independently served open-weight models. For the broader question of small versus large, it matters little which type of small model ultimately wins.

This might seem obvious—of course you shouldn't use more compute than necessary—but it runs counter to the scaling-first approach that has dominated the industry until now. Inspired by the bitter lesson, labs have heavily focused on training the most compute-intensive models possible, pushing the frontier of what AI models can achieve. With prices heavily subsidized by investors, clients had no reason to choose anything but the most advanced option. Now, as token prices rise and subsidies slow, users are facing cost pressure for the first time. It remains uncertain whether this new cost pressure will actually drive enterprise users toward smaller models. They could just as easily economize by making fewer calls, using less context, or abandoning the least promising deployments. However, if most deployments can run just as effectively on smaller models, it could significantly dampen the growing demand for inference—and raise new questions about how to justify the cost of training a frontier model.